Parameter optimization and rule base selection for fuzzy impulse filters using evolutionary algorithms

In this paper we present an effective scheme for impulse noise removal from highly corrupted images using a soft-computing approach. The filter is capable of preserving the intricate details of the image and is based on a combination of fuzzy impulse detection and restoration of corrupted pixels. In the first stage a fuzzy knowledge base required for detection of impulses as well as the optimum parameters for the fuzzy membership functions employed, is effectively 'learnt' using an evolutionary algorithm (EA). For the detection of noisy pixels and the subsequent replacement, a novel scheme where a pixel is transferred to a simulated noise free environment is introduced. We present the results for several real images and make comparisons with some of the existing noise removal methods wherever applicable to show the effectiveness of the proposed technique.